LARK-Lab/EnvFactory-1.7B
LARK-Lab's EnvFactory-1.7B is a 2 billion parameter language model, fine-tuned from Qwen3-1.7B, specifically designed for tool-use agent capabilities. It leverages an automated framework for synthesizing executable tool environments and robust reinforcement learning. This model excels at equipping LLMs with advanced tool-use functionalities, particularly for multi-turn interactions and complex agentic tasks.
Loading preview...
EnvFactory-1.7B: Scaling Tool-Use Agents
EnvFactory-1.7B, developed by LARK-Lab, is a 2 billion parameter model fine-tuned from Qwen3-1.7B. It is engineered to enhance Large Language Models (LLMs) with advanced tool-use capabilities through Agentic Reinforcement Learning (Agentic RL). The model utilizes a novel framework called EnvFactory, which automates the synthesis of executable tool environments and generates natural multi-turn trajectories for training.
Key Capabilities
- Executable Environment Synthesis: Automatically discovers, validates, and deploys tool environments based on real-world APIs.
- Topology-Aware Trajectory Sampling: Generates realistic multi-turn tool-use trajectories that capture implicit human reasoning.
- Robust RL Training: Employs verified environments and calibrated refinement for stable reinforcement learning, leading to superior performance with fewer environments (85 environments across 7 domains).
- Enhanced Tool-Use Performance: Demonstrates improved performance on multi-turn tool-use benchmarks, including BFCL Multi Turn, MCP-Atlas Pass Rate, and VitaBench Avg., compared to its base model, Qwen3-1.7B.
Good For
- Developing LLM-powered agents that require complex, stateful tool interactions.
- Applications demanding robust and scalable tool-use capabilities in dynamic environments.
- Research and development in agentic AI, particularly for synthesizing and evaluating tool-use environments.